Relevant Biophysical Parameters Discrimination along Corticospinal Tract in Patients with Stroke Using Convolutional Neural Networks

Abstract

Stroke remains the leading source of long-term disability. As the only direct descending motor pathway, the corticospinal tract (CST) is the primary pathway to innervate spinal motor neurons and one of the most well studied tracts in human neuroanatomy. Its clinical significance can be demonstrated in many distinguished traumatic situations and diseases such as stroke. Along‐tract statistics analysis enables the extraction of quantitative diffusion metrics along specific white matter fiber tracts. Besides quantitative metrics derived from classical diffusion tensor imaging (DTI), such as fractional anisotropy and diffusivities. In this study, we extracted DTI derived quantitative microstructural diffusion metrics along the CST tract in patients with moderate to severe subacute stroke. Respectively DTI metric of individual patient's fiber tract was then plotted. This approach may be useful for future studies that may compare in two different time (acute and chronic). The contribution of this work presents a totally computerized method of DTI image recognition based on conventional neural network (CNN) in order to supply quantitative appraisal of clinical characteristics. The obtained results have achieved an important classification (Accuracy=94.12%) when applying the CNN. The proposed methodology enables us to assess the classification of the used DTI images database within a reduced processing time. Experimental results prove the success of the proposed rating system for a suitable analysis of microstructural diffusion when compared to previous work.